Mark Hovde, MBA
Dan Weiner, PhD
Jean-Francois Marier, PhD

Pharsight Corporation

PK/PD scientists are using computer-assisted modeling more than ever.

Scientists conducting pharmacokinetics (PK) and pharmacodynamics (PD) studies do not have to worry about a lack of work. Candidate drugs emerging from discovery await in vivo PK evaluation prior to moving to late-stage development. Facing every-increasing clinical development costs, drug companies are performing elaborate and intensive evaluations of PK/PD at earlier stages. The US Food and Drug Administration (FDA) Critical Path Initiative is calling for more use of PK/PD modeling.

Despite rising salaries and aggressive recruiting campaigns, the industry cannot fill its demand for PK/PD scientists. Increasingly, companies are turning to technology to improve productivity and leverage scarce scientific talent.

The workflows 
Figure 1 shows the process of generating, managing, and processing data that feeds the two main outputs of PK/PD analysis—regulatory reports and other development decisions.

A protocol defines the rationale and types of subjects needed for the study, the sample size, the statistical powering, and the timing and nature of all collected information and medical procedures. A typical Phase 1 protocol will collect 40 pages of case record form information on each subject. A common sampling strategy
Basic PK/PD Workflow Schematic 
click to enlarge
Figure 1: Basic PK/PD Workflow Schematic (Source: Pharsight Corporation)
requires multiple blood samples to be collected in the first 24 hours after dosing, with more intensive sampling around key points (such as the time of maximum concentration). Clinical deviations are recorded so that any samples taken outside the time-specified scheme—or any other measurement irregularities—can be incorporated into future PK/PD analyses and reporting.

Following an analytical assay of a drug in a biological matrix, individual concentration values are merged with sample time collections. The pharmacokineticist then performs PK/PD analyses and reporting (Figure 1). The inputs of actual times and demographics from case report forms, and the involvement of statistical groups are not shown for brevity.

The production of clinical study reports that will be used to support regulatory submissions must be done in a “regulatory-compliant” fashion, with audit trails, electronic signatures, and other procedures and precautions to ensure that the final analysis and reporting conforms to FDA 21 CFR Part 11 rules.

In addition, PK/PD modelers may use data from modeling and simulation to support other decisions, such as dose optimization or go/no-go decisions. While these development decisions currently do not require production in a regulatory-compliant fashion, some organizations view that all PK/PD analyses should be performed to the higher regulatory standard.

Protocol design
Many probabilistic factors can affect the final outcome of a clinical protocol such as inaccurate estimates of PK and PD. A protocol that fails because of a bad design may condemn a good drug and tie up pharmacokineticists and modelers in unnecessary redundancy and answer-seeking. The final protocol should be optimized to ensure that the many statistical and operational design choices in the protocol interact to create a design that is likely to succeed over a reasonable range of assumptions about the PK/PD and other trial factors.

Data retrieval
The typical large pharmaceutical company performs dozens or even hundreds of trials per year, each generating time-concentration data. Ultimately, a PK/PD scientist or modeler must retrieve (often with help of a programmer) and analyze the subset of the data that contains clinical pharmacology information.

To improve data retrieval productivity, an organization must first define to the form and location for stored data. Standards for the final, analysis-ready formats for the time-concentration data sets should be adopted prior to the conduct of the trials. Many companies store the data in a PK/PD data repository—a single, secure storage medium where all clinical pharmacology information is maintained in a consistent format. PK/PD scientists can go to one place and collect data that are in a single, non-negotiable format.

How does the time-concentration data get into the PK/PD data repository in the first place? Typically, bioanalytical data reside in a laboratory information management system (LIMS). If the company has a LIMS, then a connector—built as custom code by third-party systems integrator—takes the data from the LIMS to a PK/PD data repository. Ideally, the connector will not only move data from its source into the repository, but also perform transformations to get the data into the desired analyzable format.

PK/PD Modeling, Analysis and Data Management Tools

Protocol Design
Specialized computer assisted trial design and simulation software, to test alternative protocol design assumptions and preview expected trial results.


  • EAST® (Cytel, Cambridge, Mass.)
  • Trial Simulator™ (Pharsight Corporation, Mountain View, Calif.)

Data Retrieval
PK/PD data repository, where clinical pharmacology data is securely maintained in a consistent format to enable regulatory-compliant analyses and report production.

  • Enterprise Pharmacology Series™ (ThermoFisher, Waltham, Mass.)
  • Pharsight Knowledgebase Server™ (Pharsight Corporation, Mountain View, Calif.)

Regulatory-Required Data Analysis and Reporting
Pharmacokinetic, pharmacodynamic, and noncompartmental analysis and reporting software.

  • Kinetica (ThermoFisher, Waltham, Mass.)
  • PKS Reporter™ (Pharsight Corporation, Mountain View, Calif.)
  • WinNonlin® (Pharsight Corporation, Mountain View, Calif.)
  • WinNonlin AutoPilot™ (Pharsight Corporation, Mountain View, Calif.)

Modeling and Simulation
Programming languages and related tools for building, analyzing and simulating pharmacostatistical models of drug exposure and response.

  • ADAPT-II and NPEM2 (University of Southern California)
  • NONMEM (Icon PLC, Dublin, Ireland, originally developed at the University of California, San Francisco)
  • SAS® (SAS Institute, Cary, N.C.)
  • S-PLUS® (Insightful, Seattle, Wash.)
  • WinBUGS (Institute of Public Health, Cambridge, UK)
  • Wings (University of Auckland, New Zealand)
  • Xpose (Uppsala University, Sweden)

Delivering Decisions
Software for visualizing complex, multi-dimensional PK/PD simulation results and exploring model-based drug product profiles.

  • Drug Model Explorer® (Pharsight Corporation, Mountain View, Calif.)

List compiled by Pharsight
Regulatory-required data analysis
PK parameters such as maximum concentration, total exposure (area under the curve), time to maximum concentration, half-life, and other factors are often calculated from PK data on an individual or group basis. Productivity of standardized PK analyses and reports for regulatory submission can be further optimized by scripting. Once the data set is retrieved, a standardized set of tables, analyses, and listings can be automatically generated according to defined client formats. Predefined business rules for data sets can include treatment of very small concentration values that are below the limit of quantification in the assay. Specifications for format may include fonts, text justification, treatment of subscripts and superscripts, Greek symbols, rules for table-breaking, and other factors.

Modeling and simulation
Modeling and simulation involves a set of techniques employing advanced mathematical and mechanistic (as opposed to empirical) models derived from quantitative pharmacology. The models can be used, for example, to refine a compound's dosing, a two-part procedure that predicts exposure from dose (PK modeling) and response (clinical or biomarker, indicative of toxicity or therapeutic benefit) from exposure. It may be that the response itself varies according to the state of the disease. In these cases, disease models are also useful as predictors of response in patients with more severe disease.

The FDA has a long history of applying advanced quantitative modeling. The agency's Critical Path Initiative calls for more use of PK/PD modeling and model-based drug development (MBDD) as a critical path opportunity to improve decision-making. Models can improve the prediction and assessment of patient response, and therefore probable success in the market—or earlier and less-costly failure. PK/PD modeling is usually performed at the end of Phase 1 or in early Phase 2, when there is sufficient data to build the models.

Despite the enormous strategic value they deliver to support earlier and more confident development decisions, modelers typically operate without the benefit of a regulatory-compliant workflow. The
Decision space for drug model
Figure 2: 45-Cell decision space covered by a drug model including five subpopulations, three drug attributes, and three competitors. (Source: Pharsight Corporation)  
model-building tool of choice, especially where only sparse samples are available, is the much-venerated—yet difficult to master—NONMEM (Icon PLC, Dublin, Ireland, originally developed at the University of California, San Francisco). NONMEM is a loose acronym for the statistical technique called nonlinear mixed effects modeling. Other tools such as ADAPT-II and NPEM2 (University of Southern California, Los Angeles) and WinBUGS (Institute of Public Health, Cambridge, UK) also are academic software.

Commercial NONMEM users must deal with not-so-user-friendly DOS and Fortran commands and support additional software to produce suitable commercial outputs including special software for graphics and use with parallel/grid computing environments. Commercial PK/PD scientists using these tools typically must invest in acquiring skills not just in one tool but in other software such as Xpose (Uppsala University, Sweden) and Wings (University of Auckland, New Zealand).

Once a regulatory PK analysis is produced, through automated or manual means, it must be pushed across into another work tool, usually a word processor or presentation software. From there, the reports are logged into a document repository such as Documentum (EMC Corporation, Hopkinton, Mass.).

Delivering decisions
The outputs and implications of PK/PD modeling can be complex and difficult to comprehend for those lacking a strong background in mathematics. In most cases, the modeling includes consideration of key PK-driving covariates (e.g., body weight) in the relationship between dose and all elements of the target-product profile (toxicities, also, not just dose and primary endpoint). The target-product profile may specify superior efficacy versus one competitor, and superior or equivalent toxicity versus a second competitor.

As shown in Figure 2, the total decision space may require consideration of multiple drug attributes, multiple patient sub-populations , and multiple competitors. The goal is better development decisions formed by a seamless fusion of scientific and competitive considerations. Special tools are needed to rapidly explore the large simulation space; otherwise, the team can get bogged down in the complexities.

Before regulatory documents are submitted, it is good practice to check them against the time-concentration data sets to ensure that no changes in the source data have been made or that the reports are out of sync with the raw data. A data capture system for the case record form information (paper or electronic data capture) is usually deposited into a clinical data management system. Depending on the analysis required, connectors may be needed to supply this data into the PK data repository and make it available for regulatory analysis or modeling and simulation.

Toward a more integrated modeling environment
Most managers acknowledge that PK/PD scientists waste time learning multiple analysis tools. Drug development organizations have no choice but to take proactive measures to increase PK/PD productivity. The pressure to reach surer, earlier decisions through quantitative pharmacology is certain to grow. The need for rapid, efficient production of regulatory reports is well-established.

About the Authors
Mark Hovde,Senior Vice President, Marketing, Pharsight, has experience for pharmaceutical software, data, and services companies. Dan Weiner, PhD, Chief Technology Officer, Pharsight, is an expert consultant to the FDA on pharmacokinetic modeling and bioequivalence assessment. Jean-Francois Marier, PhD, Vice President of Reporting and Analysis Services, Pharsight, has extensive expertise in a variety of studies involving PK/PD data.

This article was published in Drug Discovery & Development magazine: Vol. 10, No. 9, September, 2007, pp. 34-38.